CS551 Warm-up Project #2

Transcription

1 CS55 Warm-up Project #2 Bill Cheng

2 Multi-threading Exercise Make sure you are familiar with the pthreads library good source is the book by Nichols, Buttlar, and Farrell Pthreads Programming, O Rielly & Associates, 996 you must learn how to use mutex and condition variables correctly pthread_mutex_lock()/pthread_mutex_unlock() pthread_cond_wait()/pthread_cond_signal()/ pthread_cond_broadcast() you must learn how to handle UNIX signals pthread_sigmask()/sigwait() pthread_setcancelstate() pthread_setcanceltype() pthread_testcancel() 2

3 pthread_sigmask() #include <pthread.h> /* #include <thread.h> */ thread_t user_threadid; sigset_t new; void *handler(), interrupt(); main( int argc, char *argv[] ) { sigemptyset(&new); sigaddset(&new, SIGINT); Look at the man pages of pthread_sigmask() on nunki and try to understand the example there designate child thread to handler SIGINT parent thread blocks SIGINT pthread_sigmask(sig_block, &new, NULL); pthread_create(&user_threadid, NULL, handler, argv[]); pthread_join(user_threadid, NULL); printf("thread handler, %d exited\n",user_threadid); sleep(2); printf("main thread, %d is done\n", thr_self()); } /* end main */ 3

4 struct sigaction act; pthread_sigmask() Child thread example child thread unblocks SIGINT void * handler(char argv[]) { act.sa_handler = interrupt; sigaction(sigint, &act, NULL); pthread_sigmask(sig_unblock, &new, NULL); printf("\n Press CTRL-C to deliver SIGINT\n"); sleep(8); /* give user time to hit CTRL-C */ } void interrupt(int sig) { printf("thread %d caught signal %d\n", thr_self(), sig); } child thread is designated to handle SIGINT, no other thread will get SIGINT 4

5 Queueing Abstraction λ Q S µ S2 µ Ex: line at a bank multiprocessor executing jobs from a shared job queue ER 5

6 λ Q Arrivals & Departures C S µ S2 µ C 2 a i : arrival time d i : departure time s i : service time r i : response (system) time q i : queueing time S S2 r s d s 2 r 2 d 2 Q a a 2 0 t C C 2 6

7 Arrivals & Departures (Cont...) λ Q S µ S2 µ C C 2 C 4 C 3 S S2 r 3 s 3 d 4 d 3 s 4 r 4 q, q 2, q 3 ~ 0 q 4 > 0 q 4 Q a 3 a 4 0 t C C 2 C 3 C 4 7

8 Event Driven Simulation An event queue is a sorted list of events according to timestamps; smallest timestamp at the head of queue Object oriented: every object has a "next event" (what it will do next if there is no interference), this event is inserted into the event queue Execution: remove an event from the head of queue, "execute" the event (notify the corresponding object so it can insert the next event) Insert into the event queue according to timestamp of a new event; insertion may cause additional events to be deleted or inserted Potentially repeatable runs (if the same seed is used to initialize random number generator) 8

9 Event Driven Simulation (Cont...) Ex: 4 objects, A (arrival), Q (passive object, does not generate events), S, S2 Initially: A : [ a, create(c ) ] S : NULL Q : empty S2 : NULL only one event, next event to fire is [ a, create(c ) ] create(c ), Q->enqueue(C ) Q->dequeue(C ), S->serve(C ) A : [ a 2, create(c 2 ) ] S : [ d, destroy(c ) ] Q : empty S2 : NULL min(a 2, d ) = a 2, next event to fire is [ a 2, create(c 2 ) ] create(c 2 ), Q->enqueue(C 2 ) Q->dequeue(C 2 ), S2->serve(C 2 ) A : [ a 3, create(c 3 ) ] S : [ d, destroy(c ) ] Q : empty S2 : [ d 2, destroy(c 2 ) ] 9

10 Event Driven Simulation (Cont...) min(a 3, d, d 2 ) = d, next event to fire is [ d, destroy(c ) ] destroy(c ) A : [ a 3, create(c 3 ) ] S : NULL Q : empty S2 : [ d 2, destroy(c 2 ) ] min(a 3, d 2 ) = a 3, next event to fire is [ a 3, create(c 3 ) ] create(c 3 ), Q->enqueue(C 3 ) Q->dequeue(C 3 ), S->serve(C 3 ) A : [ a 4, create(c 4 ) ] S : [ d 3, destroy(c 3 ) ] Q : empty S2 : [ d 2, destroy(c 2 ) ] min(a 4, d 2, d 3 ) = a 4, next event to fire is [ a 4, create(c 4 ) ] create(c 4 ), Q->enqueue(C 4 ) A : [ a 5, create(c 5 ) ] S : [ d 3, destroy(c 3 ) ] Q : C 4 S2 : [ d 2, destroy(c 2 ) ] etc. 0

11 Event Driven Simulation (Cont...) C C 2 C 4 C 3 S S2 d d 2 d 4 d 3 Q a a 2 a 3 a 4 0 t C C 2 C 3 C 4

12 Time Driven Simulation Every active object is a thread a customer is a passive object, it gets passed around To execute a job for x msec, the thread sleeps for x msec nunki.usc.edu does not run a realtime OS it may not get woken up more than x msec later, and sometimes, a lot more than x msec later you need to decide if the extra delay is reasonable or it is due to a bug in your code Let your machine decide which thread to run next (irreproducible results) Compete for resources (such as Q), must use mutex 2

13 Time Driven Simulation (Cont...) You will need to implement 3 threads (or main thread and 3 child threads) the arrival thread sits in a loop sleeps for an interval, trying to match a given interarrival time (from trace or coin flip) wakes up, creates a customer object, enqueues the customer to Q, and goes back to sleep if the Q was empty before, need to signal or broadcast a queue-not-empty condition two server threads initially blocked, waiting for the queue-not-empty condition to be signaled (cont...) 3

14 Time Driven Simulation (Cont...) two server threads (cont...) when it is unblocked, if Q is not empty, dequeues a customer, sleeps for an interval matching the service time of the customer, eject the customer from the system, check if Q is empty, etc. if there is no work to perform, go wait for the queue-not-empty condition to be signaled <Cntrl+C> arrival thread will stop generating customers and terminate the arrival thread needs to clear out Q server threads must finish serving its current customer must print statistics for all customer seen 4

15 Time Driven Simulation (Cont...) Notation: α i : inter-arrival time for customer i (a i -a i- ), a 0 =0 β i : service time of customer i Initially: A : sleep(α =a ) S : idle Q : empty S2 : idle A wakes up at a : create(c ), Q->enqueue(C ) Q->dequeue(C ), S->serve(C ) A : sleep(α 2 ) S : sleep(β ) Q : empty S2 : idle A wakes up at a +α 2 : create(c 2 ), Q->enqueue(C 2 ) Q->dequeue(C 2 ), S2->serve(C 2 ) A : sleep(α 3 ) S : sleeping... Q : empty S2 : sleep(β 2 ) etc. 5

16 Coin Flipping Uniform distribution probability mass function (pmf), denoted by f(x) Probability Distribution Function (PDF), denoted by F(x) F(x) = f(x)dx = x f(w)dw f(x) = x { 0 x 0 otherwise x F(x) = 0 x < 0 x 0 x x > 6

17 Coin Flipping (Cont...) How do you flip a coin according to this distribution? Think about discrete case: f(x) bucket Add them up: f(x) f(x) x Flip a coin between 0 and 23 r r = drand48()* x 3 r lies between 3 and 3, so we have randomly chosen bucket #2 x 7

18 Coin Flipping (Cont...) Q: What were we doing when we "added them up"? A: We were doing "integration" f(x) Hint: 0 F(x) for any x x r w F(x) can numerically compute w x r = drand48() w =? 8

19 Exponential distribution Coin Flipping (Cont...) m ܵ Ñ ÑÜ Note: inter-arrival time of a Poisson process is Exponentially distributed x ܵ Ü Ýµ Ý ½ x ¼ ÑÜ r w ܵ x r = drand48() w =? 9

20 Calculating Statistics arrival thread timeout (read clock) lock & unlock stdout to print arrival msg Q try lock mutex to enter Q λ S enter Q (read clock) µ unlock mutex lock & unlock stdout to print enter queue msg try lock mutex to leave Q leave Q (read clock) unlock mutex lock & unlock stdout to print leave queue and begin service msgs begin service leave server lock & unlock stdout to print msg overhead? time in Q time in server select()? charge to no one time time between begin service and leave server is the amount of time in select() 20

21 Mean and Standard Deviation Average time for n samples, add up all the time and divide by n Average number of customer at a server same a fraction of time the server is busy 0 Average number of customer at Q time 3 2 time Standard deviation is the squareroot of variance Var[X] = E[X 2 ] - (E[X]) 2 0 2